Path: csiph.com!eternal-september.org!reader02.eternal-september.org!.POSTED!not-for-mail From: Anne & Lynn Wheeler Newsgroups: alt.folklore.computers Subject: Re: What is the oldest computer that could be used today for real work? Date: Fri, 10 Sep 2021 06:49:57 -1000 Organization: Wheeler&Wheeler Lines: 59 Message-ID: <87tuis1ikq.fsf@localhost> References: <56lljgpiq93ukr9q60fnmhsot2kl0hop35@4ax.com> Mime-Version: 1.0 Content-Type: text/plain Injection-Info: reader02.eternal-september.org; posting-host="c575e0e2593b619190b403fe7eb29729"; logging-data="13494"; mail-complaints-to="abuse@eternal-september.org"; posting-account="U2FsdGVkX18IFeSZaICfUlR1fiddKDplveYp0erN6sA=" User-Agent: Gnus/5.13 (Gnus v5.13) Emacs/27.2 (gnu/linux) Cancel-Lock: sha1:YZyGDxtm1l1OXLtSrPMlF0E1z8M= sha1:eS3+JZ+a0TREDUjpcZ2kgBOO+L0= Xref: csiph.com alt.folklore.computers:218793 Thomas Koenig writes: > Slower than the original? How many cycles per S/360 instruction does > it take? If it was really slower than the original, it would have > to be more than 100 cycles per instruction. I find that hard to > believe. Endicott cons me into helping do ECPS for 138/148. I was told that low/mid 370, 115-148 avg ten native instructions per 370 emulated instructions (i.e. 80kips 370/115 had 800kips engine, 120kips 370/125 had 1.2mips engine, etc) and the 138/148 had 6kbytes of available microcode storage. I was to identity the 6k bytes highest executed kernel instructions for moving to microcode (on roughtly byte-for-byte basis). old archived post with the analysis http://www.garlic.com/~lynn/94.html#21 6kbytes of kernel instructions pathlength accounted for 79.55% of kernel execution time ... dropped directly into microcode would run 10times faster. Also implemented for later 4331/4341. In the early 80s I got permission to give howto ECPS presentations at local user group (silicon valley) monthly baybunch meetings and would get lots of questions from the Amdahl people. They would say that IBM had started doing lots of trivial microcode implementations for the 3033 which would be required for MVS to run. Amdahl eventually responded with "macrocode" ... effectively 370-like instruction set that ran in microcode mode ... where Amdahl could implement the 3033 microcodes changes much easier with much less effort. Note the low/mid range 370s had vertical instruction microcode processors (i.e. programming like cics/risc processors). high-end 370 had horizontal microcode and typically expressed in the avg. number of machine cycles per 370 instruction. The 370/165 ran 2.1 machine cycles per 370 instruction. That was optimized for 370/168 to 1.6 machine cycles per 370 instruction. The 3033 started out as 168-3 logic remapped to 20% faster chips and the microcode was further optimized to one machine cycle per 370 instruction. It was claimed that all those 3033 microcode tweaks ran same speed as 370 (and some cases slower). Amdahl was then using macrocode to implement hypervisor support ... subset of virtual machines support w/o needing vm370 ... which took IBM several years to respond with PR/SM and LPAR in native horizontal microcode (well after 3081 and well into 3090 product life). some years later after retiring from IBM ... I was doing some stuff with http://www.funsoft.com/ and their experience with emulating 370, avg. ten instructions per 370 was about the same as low&mid range 370s ... although they had some other tweaks that could dynamically translate high-use instruction paths directly into native code on-the-fly (getting 10:1 improvement). I believe hercules is somewhat similar https://en.wikipedia.org/wiki/Hercules_(emulator) -- virtualization experience starting Jan1968, online at home since Mar1970